To improve the accuracy of image registration during feature based image stitching, especially in extreme illumination scenario, image enhancement is a crucial step as it improves the quality of the images. In this paper, a multi-scale image enhancement method is proposed as a preprocessing step of feature matching. The idea of the enhancement method is first to generate a high dynamic range version of the input images, then realize tone mapping of the HDR content by multi-scale method. SIFT descriptor and RANSAC algorithm are used for feature matching and image fusion to evaluate the proposed method. The results show that the accuracy of the feature matching can be increased by the proposed preprocessing method.